Volatility Is Distance, Not Direction

01

MOVEMENT ≠ LOSS

Make a market move.
Watch volatility refuse to judge it.

TOTAL RETURN0.0%

REALIZED VOL—

MAX DRAWDOWN—

A market can finish where it started and still travel a long, volatile road.

THE MATH

Dispersion has no preferred sign.

Standard deviation looks at how far returns spread around their average. Positive and negative surprises both enlarge it.

WHY CRASHES FEEL DIFFERENT

Equity volatility is asymmetric in practice.

Selloffs often combine large returns, changing leverage, liquidity stress, and demand for downside protection. That can lift both realized and implied volatility.

THE CORRECTION

“Volatile” does not mean “bearish.”

A smooth decline may measure low vol; a violent rally may measure high vol. Volatility is not loss, and low volatility is not safety.

02

THE EQUATION AS A MACHINE

Returns go in.
Distance comes out.

A standard historical or realized-volatility calculation does not operate on the price level. It turns a sequence of returns into one dispersion number: center, square, average, then take the square root.

01 / RETURNrₜ = ln(Pₜ / Pₜ₋₁)

Convert prices into comparable proportional moves.

02 / CENTERdₜ = rₜ − r̄

Measure each return’s distance from the sample mean.

03 / SQUAREdₜ²

Remove the sign and make large deviations count disproportionately.

04 / AVERAGEs² = Σdₜ² / (n−1)

The result is sample variance, measured in squared-return units.

05 / ROOTs = √s²

Return to the original return unit: volatility.

06 / SCALEσ_ann = s√A

Annualize with A periods only when the time-scaling assumptions are suitable.

DIRECTION-BLIND

+5 and −5 contribute equally

When they are equally far from the mean, both produce 25 after squaring. Mean return keeps direction; volatility does not.

OUTLIER-SENSITIVE

Twice the surprise, four times the variance

A deviation of 10 contributes 100; a deviation of 5 contributes 25. A few shocks can dominate a window.

SCALE-HOMOGENEOUS

Double every return, double the vol

Multiplying all returns by c multiplies volatility by |c|, while variance multiplies by c².

TRANSLATION-INVARIANT

A perfectly steady trend has zero vol

Add the same constant return to every observation and the standard deviation is unchanged. A smooth +1% every day has positive return but zero sample volatility.

SQUARING LAB

Turn one ordinary day into a shock.

Nineteen illustrative daily returns stay fixed. Move the twentieth and watch one squared deviation consume the variance budget.

FINAL-DAY RETURN+3.0%

MEAN RETURN—

Direction lives here.

DAILY VOL—

Square root of sample variance.

ANNUALIZED—

Daily vol × √252.

SHOCK’S VARIANCE SHARE—

One observation’s contribution.

DOES VOL CONTRACT TWICE AS OFTEN?

Sometimes—inside one specific update rule.

For a pure EWMA variance update,

vₜ = λvₜ₋₁ + (1−λ)εₜ²

so Δv = (1−λ)(ε²−v_old). Variance—and therefore volatility—contracts when the new absolute shock is smaller than the old volatility estimate. It expands when the shock is larger.

CONTRACTS68.3%

EXPANDS31.7%

2.15×

CONTRACTIONS PER EXPANSION

If z = ε/σ is standard normal, 68.27% of shocks lie inside ±1σ and shrink an EWMA estimate; 31.73% lie outside and enlarge it.

Conditional derivation; the decay factor λ changes the size of the update, not its direction.

Bottom line

It is not a universal law of volatility. The roughly 2:1 result belongs to an EWMA-style update under a normal standardized-shock assumption and a correctly scaled current estimate. A rolling window is approximately 50/50 under iid data; GARCH has a state-dependent threshold; implied volatility follows option prices. In Cboe’s VIX daily-close history from January 1990 through August 5, 2026, VIX fell on 53.3% of non-flat sessions and rose on 46.7%—about 1.14 contractions per expansion, not two.

03

THREE LENSES

Same unit. Different clock.

The words overlap because all three often end as an annualized standard deviation. The clean distinction is the information set and the interval being described.

01 / LOOKING BACKHV

Historical volatility

An estimate calculated from past returns over a declared window—often 20, 30, or 252 trading days.

Input: Past prices
Question: How variable has it been?
Common form: Std. dev. of daily log returns × √252

02 / PRICED NOWIV

Implied volatility

The volatility number that makes a pricing method match an option’s market price. It is forward-looking but is a price-implied, risk-neutral quantity—not a pure consensus forecast.

Input: Option prices + model
Question: What volatility is priced?
Common form: One IV per strike and expiry; often summarized

03 / OBSERVED OUTCOMERV

Realized volatility

The variability that actually occurred over a completed interval. It may use daily closes or many intraday returns.

Input: Observed returns
Question: What movement materialized?
Common form: √(sum of squared intraday returns), annualized

Important overlap

Historical volatility is usually a realized-volatility estimate applied to a trailing window. “Realized” emphasizes an observed outcome; “historical” emphasizes that the window sits behind today. Vendors may use the terms interchangeably, so inspect the formula.

04

PAST → PRICE → OUTCOME

Implied is the wager.
Realized is the reveal.

TRAILING HV18%what recently happened

IMPLIED VOL27%what the option priced

SUBSEQUENT RV12%what later happened

The option priced more movement than arrived. That does not prove the option was “wrong”: IV also embeds risk premia, supply and demand, skew, and model conventions.

PAST RETURNSHV / prior RVEstimation & forecasting input

→

TODAY’S OPTIONSIV surfaceExpectations + risk premia + positioning

→

COMPLETED WINDOWsubsequent RVThe ex-post outcome

↺

The comparison belongs on matching horizons. A 30-day IV should be compared with volatility realized over those same 30 days, with compatible annualization. The economically cleaner premium is usually discussed in variance terms: implied variance minus expected or subsequent realized variance.

05

LEVEL / PERCENTILE / RANK

One reading.
Three different statements.

Source series	Vol itself	Vol percentile	Vol rank
ImpliedOption-derived	IV = 30% The annualized volatility priced for a defined strike/expiry or summary.	IV percentile = 82 82% of comparable past IV observations were at or below 30%.	IV rank = 40 30% is 40% of the way from the lookback IV low to high.
HistoricalTrailing estimate	20-day HV = 24% Past daily returns in that rolling window annualized to 24%.	HV percentile = 68 The current rolling HV exceeds 68% of earlier rolling-HV readings.	HV rank = 31 Current HV sits 31% through its historical min–max range.
RealizedCompleted outcome	30-day RV = 19% The completed 30-day interval realized 19% annualized movement.	RV percentile = 54 That completed outcome exceeds 54% of comparable past outcomes.	RV rank = 22 It lies 22% from the lowest to highest realized window.

The formulas are reusable. “Percentile” and “rank” are transformations applied to a chosen series. They are not inherently implied-volatility concepts.

The labels are not standardized. Some platforms call a min–max rank “percentile.” Always check the vendor’s series, lookback, expiry/strike aggregation, and tie convention.

OUTLIER LAB

Make rank and percentile disagree.

Choose the underlying volatility series, set today’s level, then add one extreme observation to the lookback.

SERIES

CURRENT VOL LEVEL30%

LEVEL30%

The annualized IV reading itself.

PERCENTILE—

Counts observations below the current level.

RANK—

Uses only the lookback low and high.

PERCENTILE# observations ≤ current ÷ total × 100

RANK(current − low) ÷ (high − low) × 100

06

THE MODEL IN REVERSE

Black–Scholes prices an option.
The market price implies the vol.

EUROPEAN CALL, CONTINUOUS DIVIDEND YIELD q

C = S e^−qTN(d₁) − K e^−rTN(d₂)

d₁ = [ln(S/K) + (r − q + σ²/2)T] / (σ√T) · d₂ = d₁ − σ√T

The formula’s forward gear takes σ and returns an option value. Implied volatility uses the reverse gear:

Observe the option’s market price.
Hold the other inputs fixed.
Try values of σ until model value ≈ market price.

There is generally no simple closed-form inverse for σ, so software uses a root finder such as bisection or Newton–Raphson.

IMPLIED VOL SOLVER

Pay more for the same option.
Watch implied vol rise.

STOCK PRICE$100 STRIKE PRICE$100 DAYS TO EXPIRY30 CALL MARKET PRICE$4.00

Illustration assumes a European call, 4% continuously compounded risk-free rate, no dividends, frictionless trading, and a single constant volatility.

THE σ THAT CLOSES THE GAP

— IMPLIED VOLATILITY

MODEL PRICE—

PRICING GAP—

1-DAY VOL SCALE—

EXPIRY VOL SCALE—

Move the option price to run the model backward.

What the number is—and is not

Black–Scholes–Merton assumes a simplified world, including continuous paths and constant volatility. Real option markets produce a surface: IV varies by strike and expiry. American exercise, discrete dividends, jumps, rates, borrow, bid–ask spreads, and stale quotes can require different pricing methods or careful adjustments.

07

OTHER WAYS TO GET A VOL NUMBER

Measure, forecast, or imply.
Do not mix the jobs.

A / OBSERVE

Past & realized measures

Simple and common: calculate log returns, subtract their mean, estimate standard deviation, then annualize—often with √252 for daily equity data.

Use the intraday high–low range, and sometimes open and close. They can use more price-path information, but each carries assumptions about drift, gaps, and trading continuity.

Samples many returns inside the completed period. More observations can reveal the path, but market microstructure noise, jumps, and sampling frequency matter.

B / FORECAST

Conditional models

Transparent and easy to explain, but a return’s weight drops from full to zero the moment it leaves the window.

Exponential weights let shocks fade gradually. The decay parameter controls how quickly the model forgets.

These frameworks acknowledge that volatility changes through time and tends to cluster. Their output is a conditional forecast, not observed volatility.

C / INFER

Option-derived measures

Each liquid strike and expiry can imply a different volatility. A single “stock IV” requires a vendor’s aggregation rule.

A smooth surface respects the smile or skew the market actually quotes. Local-volatility and stochastic-volatility models describe that surface differently.

VIX-style methods combine many out-of-the-money calls and puts to estimate expected variance over a target horizon instead of averaging Black–Scholes IVs.

ANNUALIZATION IS A UNIT CONVERSION, NOT A FORECAST

If daily volatility is 1.25%, the familiar square-root-of-time conversion gives roughly 1.25% × √252 ≈ 19.8% annualized. That scaling assumes returns behave suitably across time; autocorrelation, jumps, changing regimes, and calendar choices can break the shortcut.

08

A SHORT HISTORY OF AN INVISIBLE QUANTITY

Volatility began as a model.
Then markets learned to quote it.

1900

RANDOM MOTION

Louis Bachelier

Théorie de la spéculation used a mathematical random process to study price changes and option-like claims—years before Brownian motion became standard financial language.

1952

RISK AS DISPERSION

Harry Markowitz

Mean–variance portfolio theory helped establish variance and standard deviation as operational measures of investment risk, while covariance connected assets inside a portfolio.

1963

FAT TAILS

Benoit Mandelbrot

Observed speculative-price changes challenged the comfortable Gaussian picture: large moves occurred more often than a simple normal model suggested.

1973

VOL INSIDE A PRICE

Black, Scholes & Merton

Black and Scholes published their option-pricing formula; Merton developed and generalized the continuous-time framework. Once option prices trade, the volatility input can be inferred backward.

1980–82

BETTER MEASUREMENT, CHANGING VARIANCE

Ranges & ARCH

Parkinson and Garman–Klass extracted information from daily ranges and OHLC prices. Robert Engle’s ARCH model made conditional variance explicitly time-varying.

1993 / 2003

VOL AS AN INDEX

VIX, then a new VIX

Cboe introduced VIX in 1993. In 2003 it replaced the original at-the-money Black–Scholes-IV approach with a model-free method using a broad strip of SPX options.

2000s →

THE PATH IN HIGH RESOLUTION

Realized-volatility research

High-frequency data and quadratic-variation theory connected intraday squared returns to practical realized measures, forecasts, risk systems, and variance markets.

09

PUTTING THE PIECES TOGETHER

Four patterns worth recognizing.

EVENT VOL

IV can rise before earnings—then collapse after the announcement.

Uncertainty is concentrated before the event. Even a large price gap can be followed by lower IV once the range of possible outcomes narrows. That is “vol crush,” not proof that the event was small.

THE RISK PREMIUM

IV often sits above subsequent RV.

Option buyers may pay to transfer convex or crash risk, while sellers demand compensation. But the spread changes through time and is not free money; rare shocks can dominate many quiet expiries.

SURFACE, NOT DOT

There is no single universal IV for a stock.

Different strikes and expiries imply different vols. “IV = 30%” is incomplete until you know which option or exactly how a platform summarized the surface.

CONTEXT IS LOCAL

30% can be high for one asset and low for another.

The level tells magnitude. Percentile and rank compare that magnitude with the asset’s own history. Neither tells you direction, fair value, or which strategy will win.

BEFORE YOU TRUST A VOL NUMBER

Ask for its coordinates.

01
What underlying? Index, single stock, rates, crypto, and commodities live on different scales.
02
Which horizon? 1-day, 20-day, 30-calendar-day, or one-year?
03
Which estimator? Closes, OHLC, intraday returns, model forecast, or options?
04
Which annualization? Trading days or calendar days, and is square-root scaling defensible?
05
Which comparison set? Lookback, sampling frequency, IV surface summary, and percentile/rank convention?

10

SOURCES & METHOD NOTES

Read past the infographic.

1900 · ORIGINAL PAPERBachelier, “Théorie de la spéculation”↗ 1952 · ORIGINAL PAPERMarkowitz, “Portfolio Selection”↗ 1963 · ORIGINAL PAPERMandelbrot, “The Variation of Certain Speculative Prices”↗ 1973 · ORIGINAL PAPERBlack & Scholes, “The Pricing of Options and Corporate Liabilities”↗ 1997 · HISTORICAL CONTEXTRoyal Swedish Academy: the Black–Scholes–Merton contribution↗ OCC / OIC · DEFINITIONSHistorical and implied volatility↗ OCC / OIC · METRICSIV rank and IV percentile formulas↗ CBOE · METHODOLOGYWhat VIX measures and how its horizon is held near 30 days↗ CBOE · HISTORYOld VIX, model-free VIX, and maturity interpolation↗ CME · TECHNICAL NOTEVolatility estimation and square-root-of-time scaling limits↗ 1980 · ORIGINAL RESEARCHGarman & Klass on OHLC volatility estimators↗ 1982 · ORIGINAL PAPEREngle introduces ARCH↗ 2003 · ORIGINAL RESEARCHAndersen et al. on high-frequency realized volatility↗ NIST · NORMAL DISTRIBUTION68.27% of a normal distribution lies within ±1 standard deviation↗ 1996 · RISKMEASUREMENTRiskMetrics technical document and EWMA methodology↗ CBOE · DAILY DATAVIX history used for the dated contraction/expansion check↗

Educational illustration only—not investment advice, a trading signal, or a representation of any live security. Lab data are synthetic. Percentile/rank examples use transparent simplified conventions; market-data platforms can differ.

BACK TO THE LIBRARYChoose another rabbit hole →

Volatility is distance, not direction.

Make a market move.Watch volatility refuse to judge it.